[Bug ld/32624] broken heuristics in R_386_GOT32{,X} handling

[jbeulich at suse dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=32624

--- Comment #15 from Jan Beulich <jbeulich at suse dot com> ---
(In reply to H.J. Lu from comment #13)
> Do you truly believe that R_386_GOT32 is mishandled by ld?

Why "believe"? You've demonstrated it (part of the problems, that is) to
yourself and everyone else in comment 5. To me that's rather "I know it's
broken".

(In reply to H.J. Lu from comment #14)
> "lea foo@GOT, %ecx" is a special case, see PR 21168.  Maybe assembler should
> disallow foo@GOT for instructions with the last byte of opcode == 0x8d.

Well, no. First and foremost because that won't help. Just to repeat: The
linker needs to work correctly on (correct) object files of any origin. Plus
using .reloc  one can construct about anything even with gas.

As to an object file being "correct" when it has what we see here: See the very
first question in the original description. If what you suggest was spelled out
anywhere, rejecting this already in gas would be an option. Yet (I'll be happy
to be proven wrong) I can't help the impression that none of this is part of
any specification whatsoever.

-- 
You are receiving this mail because:
You are on the CC list for the bug.